Parameterization, regionalization and radiative transfer coherence of optical measurements acquired in the St-Lawrence ecosystem

(1)

Departement de Geographie et Teledetection Faculte des lettres et sciences humaines

Universite de Sherbrooke

Proprietes optiques intrinseques et apparentes

des eaux du golfe et de l'estuaire du

Saint-Laurent: concordance optique,

parametrisation et variabilite spatio-temporelle

Parameterization, regionalization and radiative

transfer coherence of optical measurements

acquired in the St-Lawrence ecosystem

Servet Ahmet Qizmeli

These presentee pour l'obtention du grade de Philosophise Doctor (Ph.D.) en teledetection

(2)

**1*1**

Library and Archives Canada Published Heritage Branch 395 Wellington Street Ottawa ON K1A0N4 Canada Bibliotheque et Archives Canada Direction du Patrimoine de I'edition 395, rue Wellington Ottawa ON K1A0N4 Canada

Your file Votre reference ISBN: 978-0-494-37961-5 Our file Notre reference ISBN: 978-0-494-37961-5

NOTICE:

The author has granted a non-exclusive license allowing Library and Archives Canada to reproduce, publish, archive, preserve, conserve, communicate to the public by

telecommunication or on the Internet, loan, distribute and sell theses

worldwide, for commercial or non-commercial purposes, in microform, paper, electronic and/or any other formats.

AVIS:

L'auteur a accorde une licence non exclusive permettant a la Bibliotheque et Archives Canada de reproduire, publier, archiver,

sauvegarder, conserver, transmettre au public par telecommunication ou par Plntemet, prefer, distribuer et vendre des theses partout dans le monde, a des fins commerciales ou autres, sur support microforme, papier, electronique et/ou autres formats.

The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission.

L'auteur conserve la propriete du droit d'auteur et des droits moraux qui protege cette these. Ni la these ni des extraits substantiels de celle-ci ne doivent etre imprimes ou autrement reproduits sans son autorisation.

In compliance with the Canadian Privacy Act some supporting forms may have been removed from this thesis.

Conformement a la loi canadienne sur la protection de la vie privee, quelques formulaires secondaires ont ete enleves de cette these. While these forms may be included

in the document page count, their removal does not represent any loss of content from the thesis.

Canada

Bien que ces formulaires

aient inclus dans la pagination, il n'y aura aucun contenu manquant.

(3)

1

Cette these a ete evaluee par un jury compose des personnes suivantes :

M. Norm T. O'Neill, directeur de recherche (Departement de Geographie et Teledetection, Universite de Sherbrooke)

Mme. Suzanne Roy, codirecteur de recherche (ISMER)

M. Alain Royer, examinateur interne (CARTEL, Universite de Sherbrooke)

M. Hardy Granberg, examinateur interne (CARTEL, Universite de Sherbrooke)

M. Marcel Babin, examinateur externe (Laboratoire d'Oceanographie de Villefranche, Universite Pierre et Marie Curie)

(4)

11

Resume

Les proprietes bio-optiques et bio-physiques du golfe et de l'estuaire du Saint-Laurent ont ete evaluees au cours de 5 croisieres multi-saisonnieres menees entre 1997 et 2001. Les proprietes optiques intrinseques (POI) mesurees comportaient des proflis verticaux du coefficient d'absorption et d'attenuation et la fonction de diffusion volumetrique ainsi que le coefficient d'absorption par les particules (total), par les particules non-algales, par le phytoplancton et par la matiere organique dissoute (CDOM). Les parametres optiques apparents (POA) incluaient les profils verticaux de la luminance ascendante et de l'eclairement descendant. L'allure spectrale des principaux POI comme l'absorption par le phytoplancton, le CDOM et les particules non-alguales comme la retrodiffusion particulaire ont ete parametrises en utilisant les modeles conventionnels mais aussi des adaptations de ces modeles. Les statistiques descriptives de chaque variable de cette collecte de donnees sont decrites. Les variations spatiales et temporelles observees dans les parametres biophysiques et leurs allures spectrales sont egalement presentees. Les resultats indiquent que les taux mesures de POI et de POA varient selon des proportions deja observees dans des ecosystemes cotiers semblables. Les patrons de variation spatiale et temporelle etaient, en general, conformes aux variations attendues vis-a-vis des considerations theoriques et experimentales pour cet ecosysteme particulier. D'ailleurs, les allures spectrales des POI et des POA se sont deelarees conformes avec la variation detectee dans les concentrations (visiblement signicatives) des constituants mesures.

Une serie complete des sections efficaces des POI a ete calculee a partir des donnees acquises en mer. La variation des sections efficaces a ete conforme aux taux de variation deja decrits dans la litterature. Ces resultats nous permettent de conclure que le jeu de donnees que nous possedons pourrait etre utilise dans le calcul de la concentration des constituants bio-geochimiques a partir des POI associes. Une etude de concordance des donnees bio-optiques au niveau de transfert radiatif a ete effectuee. Les parametres optiques simules correspondaient dans un meme ordre de grandeur avec les parametres optiques mesures. Ce resultat nous a permis de confirmer la validite de notre jeu de donnees acquis en mer avec une approche theorique.

Nous pouvons avancer le fait que cette campagne de collecte de donnees a permis de decrire a une large echelle le comportement des regions bio-optiques du Saint-Laurent dans le contexte des objectifs de la teledetection. Cet ensemble de donnees bio-optiques fournit une base solide pour le developpement d'un modele rigoureux d'inversion de la concentration de Chla a l'aide d'un algorithme satellitaire pour les eaux de type II du systeme du Saint-Laurent.

Mots-clefs:

phytoplacton, parametres bio-optiques, matiere organique dissoute, parametres optiques apparents, proprietes optiques intrinseques, trasfert radiatif, eaux de type II, matiere en suspension.

(5)

Ill

Abstract

In-water biogeochemical constituents and bio-optical properties of the St-Lawrence Gulf and Estuary were monitored during 5 cruises conducted between 1997-2001 accross different seasons. Measured inherent optical properties (IOPs) included vertical profiles of the absorption and attenuation coefficients and the volume scattering function as well as absorption by particles, non-algal particles, phytoplankton and coloured dissolved organic matter (CDOM). Apparent Optical parameters (AOPs) included vertical profiles of the upwelling radiance and downwelling irradiance. The spectral shape of the major IOPs like absorption by phytoplankton, CDOM and non-algal particles as well as the particulate backscattering were parameterized using conventional models and adaptations of conventional models. Descriptive statistics of each variable in the collected dataset were analysed and compared with previous findings in the literature. The optical coherence of the measurements was verified using a radiative transfer closure approach. A complete set of IOP cross-sections for optically significant biogeochemical variables were generated.

The magnitude and the spatial, temporal and spectral variation exhibited by the optically significant inwater biogeochemical constituents as well as the bio-optical parameters was consistent with our current knowledge of the ecosystem. The variation of the bio-optical parameters throughout the seasons was also coherent with our expectations. All the measured and derived parameters were found to vary within the ranges reported in the literature. Evidence was presented wherein the Gulf waters, which are usually considered as case I waters could also behave like case II waters. Moreover, spectral signatures exhibited by the IOPs and AOPs were coherent with the variation detected in the concentrations of the measured (optically significant) constituents. The extracted IOP crosssections were consistent with the results of similar studies previously performed and could eventually be used in the estimation of the biogeochemical constituent concentrations given the related component IOPs. First-order radiative transfer closure was achieved; this underscored the validity of our experimental dataset based on considerations of higher level, integrative, physics.

We argue that the current data collection campaign succeeded as a comprehensive framework for describing the behavior of the St-Lawrence bio-optical provinces within the context of remote sensing objectives. This bio-optical dataset should provide the basis for the development of a

rigorous, satellite-based, remote sensing algorithm for the retrieval of near surface chlorophyll,

fine-tuned to the local characteristics of the St-Lawrence system.

Keywords:

phytoplankcton, bio-optical parameter, coloured dissolved organic matter, Apparent Optical parameter, inherent optical properties, radiative transfert, case II waters, suspended matter.

(6)

IV

Remerciements

J'aimerais tout d'abord remercier mes supervisees Norm O'Neill et Suzanne Roy pour leur patience infinie avec moi. Un grand merci a Shems-Eddine Zidane. Sans toi, 9a n'aurait pas ete possible. Merci Mehrnet d'avoir toujours ete la. Merci a toutes et a tous dont je ne peux citer le nom ici. Enfvn mes parents et ma soeur... Ce cadeau (tardif) est dedie a vous.

(7)

List of Figures

3.1 The main geographical features of the St. Lawrence ecosystem 33

3.2 The four distinct primary production zones of the lower St. Lawrence estuary 35

3.3 Locations of all the stations that were visited during the 1997-2001 cruises. . 36

3.4 The effects of the ac-9 drift correction on deionized water absorption spectra 40

3.5 The effects of the ac-9 temperature correction on deionized water

absorp-tion spectra 42

3.6 The effects of the ac-9 scattering correction on deionized water absorption

spectra 43

3.7 Exponential fit to spectral a^AP data 46

3.8 ECO-VSF3 dark current measurements 49

4.1 Spatial distribution of vertically averaged fluorimetric Chla concentrations . 55

4.2 Temporal variability of vertically averaged values of Chla concentration . . 57

4.3 Spatial distribution of vertically averaged SPM concentrations 59

(12)

4.5 Spatial distribution of vertically averaged POM concentrations 61

4.6 Spatial distribution of vertically averaged %POM 62

4.7 Temporal variability of vertically averaged values of Suspended Matter . . 63

4.8 Temporal variability of vertically averaged values of Particulate Inorganic

Matter PIM 64

4.9 Temporal variability of vertically averaged POM 65

4.10 Temporal variability of vertically averaged %POM 66

4.11 Sample spectra of the vertically integrated absorption components

mea-sured with the ac-9 69

4.12 Components of absorption spectra measured with the filter-pad method . . 72

4.13 Spatial distribution of the absorption components as determined by the

filter-pad method. The April 2001 dataset 73

4.14 Spatial distribution of the absorption components as determined by the

filter-pad method. The May 2000 dataset 74

4.15 Spatial distribution of vertically integrated a^AP (440) 75

4.16 Spectrophotometric CDOM absorption spectra 77

4.17 Spatial distribution of vertically averaged ag 78

4.18 Spatial distribution of surface Sg 79

4.19 Spectra of the scattering coefficient b(A) 80

4.20 Spatial distribution of vertically averaged b\, and b\, 83

(13)

4.21 Spatial distribution of the vertically averaged mass-specific scattering

coef-ficient b*p (550) 84

4.22 Spectra of single scattering albedo 85

4.23 Spatial distribution of the vertically averaged single scattering albedo, April

2001 86

4.24 Spatial distribution of the single scattering albedo, April 2001 87

4.25 Remote-sensing reflectance Rrs and diffuse attenuation coefficient iQ spectra 90

5.1 Scatterplots of aNAP(MQ) and 0^(440) with respect to Salinity, SPM and PIM 93

5.2 Scatterplotof S^anda|(375) 96

5.3 Scatterplots of aNAP(U0) and ap(440) with respect to PIM and SPM . . . . 100

5.4 Scatterplots of a*NAp with aph and %POM 101

5.5 Scatterplots of a^ and a*h versus Chla 103

5.6 (a) Simulated versus measured Rrs values for four wavelengths, r2=0.93 (all

data altogether). All valid data pairs from April 2001 and May 2000 were shown, (b) Spectra of the mean relative error between all the valid simu-lated and measured Rrs(A). The vertical lines show the standard deviation

of the relative error. 108

5.7 Simulated and measured Rrs, by and b\,/b spectra I l l

5.8 Simulated and measured Rrs, b\, and by/b spectra 112

5.9 Simulated versus measured surface by for the two measurement wavelengths, r2-0.91 (all data altogether). All valid data pairs from April 2001 are shown. 113

(14)

5.10 Results of the correlation analysis between the spectral backscattering frac-tion bfc(A) and the spectral slope of the scattering coefficient Y-p. Variafrac-tion of r2 with respect to wavelength (a); scatterplot of fob(530) with respect to

Yp(b). April 2001 and May 2000 datasets pooled together 114

6.1 Scatterplots of actual measurements of ag, Chla and SPM 116

6.2 Spatial distribution of the ratio ag(440)/Chla 118

6.3 Spatial distribution of the ratio SPM/Chla 119

6.4 Spatial distribution of the ratio ag(440) /SPM 120

6.5 Spectra of the ratio of the absorption components to the total non-water

absorption 124

6.6 Spatial distribution of flp/j/anu;/ agl'anw and aj^Ap/anw for the 2001 cruise. . . 125 6.7 Spatial distribution of a^ la

nwi an I anw and a^AP /&nw for the 2000 cruise. . . 126

6.8 Ternary plots of the absorption budget for the 2001 cruise 128

6.9 Ternary plots of the absorption budget for the May 2000 cruise 129

(15)

List of Tables

3.1 Inherent optical parameters measured at the Estuary and Gulf of St.Lawrence 36

4.1 Statistics of in-water constituent concentrations 56

4.2 Absorption component statistics 67

4.3 Statistics of the extensive IOPs 68

4.4 CDOM absorption statistics 76

4.5 Statistics of scattering parameters 81

5.1 Sg compared with values previously published in the literature 97

5.2 SNAP compared with values previously published in the literature 98

(16)

Nomenclature

%POM Fraction of organic matter, percents ft Volume scattering function, m~x

ftw Volume scattering function by water molecules, m_ 1

A Wavelength, nm O Solid angle, steradians

co Single scattering albedo, dimensionless

cop Single scattering albedo by particles, dimensionless (p Azimuth angle, degrees

p Fresnel reflectance, dimensionless T Optical depth, dimensionless 6 Zenith angle, degrees

6 Zenith angle of refracted light, degrees f> Scattering phase function, dimensionless

hi The total (bulk) backscattering fraction, dimensionless bp Backscattering fraction by particles, dimensionless

(17)

a Total (bulk) absorption coefficient, m_ 1

a^AP Mass-specific absorption coefficient by non-algal particles, m2.g~l

ag Absorption coefficient by CDOM, m~l

flp Absorption coefficient by total particles, m_ 1

aw Absorption coefficient water molecules, m~l

a^„ Absorption coefficient by non-algal particles (detritus) and CDOM (gelbstoff), m_ 1

a^AP Absorption coefficient by non-algal particles, m~l

anw Non-water absorption coefficient (a-aw), m"1

dph Absorption coefficient by phytoplankton, m~l

a*h Mass-specific absorption coefficient by phytoplankton, m2.g_ 1

AOP Apparent optical properties

b Total (bulk) scattering coefficient, m_ 1

b* Mass-specific particulate scattering coefficient, m2.g~~x

b* Mass-specific particulate scattering coefficient, m2.g~l

b\, Total (bulk) backscattering coefficient, m_ 1

bp Scattering coefficient by particles, m_ 1

bw Scattering coefficient by water molecules, m_ 1

b\)V Backscattering coefficient by particles, m_ 1

b\)W Backscattering coefficient by water molecules, m~l.sr~l

(18)

cnw Attenuation coefficient by everything except water molecules, m"1

CDOM Chromophoric (colored) dissolved organic matter Chla Chlorophyll-a concentration, mg.m~3

CV Coefficient of variation, percents E Total (spherical) irradiance W.m~2

Erf Downwelling irradiance W.m~2

Es Downwelling irradiance in air, W.m2

Eu Up welling irradiance W.m~2

IOP Inherent optical properties

K Spectral slope parameter of backscattering fraction, dimensionless Kd Diffuse attenuation coefficient for downwelling irradiance, m~l

L Total (spherical) radiance W.m_ 2.sr_ 1

L^ Downwelling radiance W'.m~2.sr^1

Lu Upwelling radiance ]N.m~2.sr~l

Lw Water-leaving radiance W.m~2.sr~l

n Index of refraction of water relative to air, dimensionless Phaeo Phaeopigment concentration, mg.m~3

PIM Particulate inorganic matter concentration, mg.L-1

POM Particulate organic matter concentration, mg.L~l

R Reflectance, dimensionless

(19)

Rrs Remote sensing reflectance, sr l

S Salintiy, PSU

Sg Spectral slope of agr nm~l

SNAP Spectral slope of a^APr nm-1

SPM Suspended sediment concentration, m g . L- 1

std Standard deviation

Yp Spectral slope of particulate scattering coefficient, dimensionless z Water depth, m

(20)

Chapter 1 Introduction

1.1 The problem of remote sensing of coastal waters

It is possible to estimate concentrations of chlorophyll (Chla), suspended matter (SPM) and chromophorous dissolved organic matter (CDOM) in the upper layer of the water bodies using remotely sensed color data using simple empirical reflectance ratios (Gordon and Clark, 1980). These algorithms are expected to accurately transform the changes in the reflectance spectra to the changes in the concentration of the constituent of interest. This transformation process is achieved by making certain assumptions about the way remote sensing reflectance spectra responds to the changes in the concentration of the optically significant constituents.

Satellite-based operational retrieval of Chla is currently being achieved by band-ratio al-gorithms over the open ocean (Gordon and Clark, 1980). The fundamental assumption of these algorithms is that phytoplankton is the main optically significant in-water com-ponent affecting the apparent reflectance curve. Although providing satisfactory results in the open ocean (O'Reilly et al., 1998a), these algorithms generally fail in coastal regions where non-chlorophyllous components, namely the CDOM and SPM are also present

(21)

in the water. When they are of terrestrial origin, these constituents do not covary with Chla. This induces independent and significantly different spectral responses of the water medium (and thus water color) hence confusing the traditional 2-band ratio algorithms. Water bodies with these kinds of (turbid) optical characteristics are usually found near the coasts, inland seas, rivers and lakes. This type of water is historically categorized as case II in contrast to case I (open ocean) waters (Morel and Prieur, 1977). The currently op-erational SeaWiFS case-I (open-ocean) algorithm OC4v4 (O'Reilly et ah, 1998b) is known to overestimate the chlorophyll concentration over our study site, the St. Lawrence, se-riously limiting the quantitative use of the currently available satellite ocean color data (Yayla et ah, 2006). In addition to the problems defined above, remote sensing of the op-tically complex waters of the St.Lawrence Estuary and Gulf represents additional chal-lenges that are due to the extremely high spatial and temporal variability of its optical characteristics (Cizmeli, 2000, Jacques et ah, 1998). Successful inversion of satellite-based observations of water quality parameters over coastal waters is thus problematic because of our inadequate knowledge of the inherent optical properties (IOP) of the constituents present in water and their respective relationship with the apparent optical properties (AOP) measured above water (Babin et ah, 2003b). See chapter 2 for a more detailed de-scription of IOPs and AOPs.

1.1.1 Inverse IOP models for the retrieval of Chla

When dealing with the retrieval problem of remote sensing, one needs to estimate the IOPs from the measured AOPs (the inverse approach). The number of degrees of freedom in the AOP measurements (i.e. the number of bands) and the degree of independence between the optical contributions of each constituent controls the retrieval separability of individual IOPs; there is usually more than one IOP combination that would result in the measured AOP field. For the inverse approach, one thus needs to make some assump-tions on the IOPs from a priori knowledge. This process is usually called inverse modeling.

(22)

There are many different types of inverse IOP models that have recently been developed (Loisel and Stramski, 2000, Hoge and Lyon, 1996, Carder et ah, 1999, Garver and Siegel, 1997, Hoge and Lyon, 1996, Lee et ah, 2002). IOCCG (2006) gives a comparison of 9 inverse models that include but are not limited to the models cited just above. In their study, the authors apply all the considered models on the same sets of synthetic and in-situ datasets. All these models use different satellite channels and make different assumptions on IOPs, hence producing different results. Most of these models take Rrs(A) or R(A) as input and provide one or more of the following IOPs: the total absorption a (A) and total backscatter-ing b&(A) coefficients as well as the phytoplankton absorption ap/j(A) and absorption due

to combined detritus (by-products of biological activity) and CDOM, fl^(A). The authors conclude that the models perform better for clearer waters (a(440) < 0.3 m~l) and that the bulk coefficients a (A) and &&(A) were the ones which were retrieved with the highest ac-curacy. Less reliable results were obtained when decomposing the absorption coefficient into its components a^ and a^g, partly due to the overlapping absorption signatures of these two components which are difficult to estimate using multispectral Rrs data using a subset of SeaWiFS channels only. The same study recommends that the scientific commu-nity should start considering a (A) and &&(A) as standard ocean color products and that more effort should be spent on the decomposition of the absorption components as newer methods for the decomposition of the bulk IOPs (Lee etal, 2002) as well as improved spec-tral measurements of the IOP component ratios (Babin et ah, 2003b) have recently started to become available. The reader is referred to IOCCG (2006) for more details.

(23)

1.2 Brief overview of bio-optical oceanography of case II

waters

With the new generation of in-situ IOP-meters that became available over the last few years, the bio-optical oceanographers started the acquisition of extensive sets of IOP data with an unprecedented spatial, temporal and spectral coverage (Bowers and Binding, 2006, Binding et al, 2003, Doxaran et al, 2004, 2006, Babin et al, 2003b, Vantrepotte et al, 2007). Among these, there are still very few studies that priorise the seasonal variations of the IOPs (Vantrepotte et al, 2007). Although considerable knowledge have been acquired with the new data, the complete picture is still far from being available for a successful inversion of the remote sensing signal over optically complex coastal waters. As dis-cussed in the previous section, retrieval of Chla and the other in-water constituents from remotely sensed data over case II waters require a generalization of the spectral, temporal and spatial variation of the IOPs that are fine-tuned to the local characteristics of the stud-ied ecosystem. Inverse IOP models (Gallegos et al, 1990, Roesler and Perry, 1995, Sydor et al, 1998, Barnard et al, 1999, Stramska et al, 2000) have been used by the scientific com-munity since at least two decades to achieve a better understanding of such variations. Most of these models were either semi-empirical or analytical and were calibrated using in-situ datasets of the absorption and scattering coefficients. Information about backscat-tering, on the other hand, was either being assumed as spatially and vertically invariable, or measured at a single wavelength and at discrete depths. Use of new-generation IOP datasets that better describe the spectral characteristics of the backscattering coefficient (and of the volume scattering coefficient at more angles) over the entire visible spectrum would certainly improve the performance of the upcoming case II satellite algorithms considerably.

Researchers have traditionally been performing radiative transfer simulations not only to build IOP models, but also to verify the coherence of newly acquired experimental

(24)

datasets (Barnard et ah, 1999) usually acquired with new (and rather unknown) instru-ments. This latter approach involves a radiative transfer closure analysis; a procedure based on the energy conservation principle that is applied through the comparison of experimental and analytically computed optical parameters. Recent radiative transfer closure studies have started to incorporate in their simulations backscattering measure-ments acquired at a single wavelength (Chang et ah, 2003, McKee et ah, 2003). It has also been found in some more recent studies that the incorporation of the spectral shape of the backscattering coefficient at one, two or six channels offers the potential of considerably improving the performance of the closure (Bulgarelli et ah, 2003, McKee and Cunningham, 2005, Tzortziou et ah, 2006). To our knowledge, very few radiative transfer studies have been performed using the full spectral characteristics of the backscattering coefficient.

There exists very few studies on the inherent optical properties (see section 2.1) of the St.Lawrence ecosystem. Babin et ah (1993) were the first to publish chlorophyll-specific absorption spectra for the Gulf and the Estuary and found that choosing one constant specific-absorption spectra to describe the whole ecosystem could result in errors u p to 24% in the primary productivity estimations. Nieke et ah (1997) measured the absorption and fluorescence of CDOM along a transect crossing the Estuary and Gulf and found a strong inverse correlation between CDOM absorption coefficient ag and salinity. Roy et ah (2003) analyzed the absorption components obtained with the filter-pad method collected during May 2000 and April 2001 cruises of the current study. Significant differences were found between the optical characteristics of the phytoplankton assemblages from one re-gion to another and between the two years. The authors also analyzed the chlorophyll-specific absorption spectra along with HPLC-derived (High Performance Liquid Chro-matography) pigment information in order to explain the influences of the packaging ef-fect. See section 5.3 for a more detailed explanation of the results they found.

Timeseries of the CZCS satellite sensor acquired over the Gulf of St. Lawrence over the 1979-1981 period were analyzed by (Fuentes-Yaco et ah, 1997b). The authors observed that

(25)

the average seasonal pigment concentrations as estimated from CZCS images were higher in early-fall, especially in the Anticosti Gyre and at the south-west of the Anticosti Island than in spring where no substantial bloom was apparent in their imagery. The authors concluded that the spatial patterns of the satellite signal visible in the Gulf were coherent with their knowledge of the wind and buoyancy forcings usually dominant in the region (Fuentes-Yaco et al., 1997a) and linked the phenomena of early-fall blooms to the mixing patterns that are more often dominated in fall by the winds than by the freshwater input from the Estuary and other sources as these latter decreased in fall.

The data presented in this thesis are part of a multi-year study where the in-water con-stituent concentrations, in-water bio-optical properties and in-water, above water, air-borne and spaceair-borne remotely sensed optical data were acquired over the Gulf and Es-tuary (Larouche, 1998, 2000). These studies briefly overviewed the in-water radiometric data that we employed in the current work. Jacques et al. (1998) used above-water ra-diometric spectra and airborne imaging spectrometry in order to evaluate the correlation between the simple and multi-band ratios and surface chlorophyll concentration. The au-thors found better correlations for the signal normalized by the 670 nm channel. Cizmeli (2000) attempted to estimate the intra-pixel variability of the SeaWiFS images using var-iograms applied to low altitude hyperspectral imagery, finding that in the highly patchy coastal waters of the St. Lawrence, the amount of spatial detail that was being averaged out within a SeaWiFS pixel was higher than the maximum intra-pixel variance employed as a design target in the SeaWiFS project (35% of the overall variability). Silio-Calzada (2002) estimated the ratio of total backscattering coefficient to the bulk absorption coef-ficient (bb/a) of the Estuarine waters using airborne hyperspectral imagery. The author used the model of Jerome et al. (1996) and was able to make a classification of the prin-cipal optically-active water components present in the water column using b^/a at three different wavelengths.

(26)

1.3 Objectives

The St.Lawrence bio-optics project is a joint research initiative funded by the Canadian governmental institutes CRSNG, MPO and FQRNT. The research is conducted by the re-mote sensing laboratory of Maurice-Lamontagne Institute (MLI), Centre d'Applications et de Recherche en Teledetection (CARTEL) of Universite de Sherbrooke, Institut de Sci-ences de Mer (ISMER) of Universite de Quebec a Rimouski (UQAR) and York University. The general aim of the project is to develop a case II satellite chlorophyll extraction al-gorithm fine-tuned to the particular characteristics of the region (Therriault et ah, 1993). The current study aims to help this by achieving a better understanding of the relation-ships between the optically significant biogeochemical constituents and the IOPs. More specifically, the two main objectives are :

1. to quantify the relative contribution of each IOP component to the bulk IOP;

2. to express the local (St. Lawrence) variations in the optically active biogeochemical constituents as a function of IOPs;

Realization of the first objective will allow us to estimate constituent IOPs from bulk IOPs that were derived from inverse models (described in section 1.1). The second objective will make it possible to infer the constituents concentrations once constituent IOPs are retrieved from ocean color imagery (the previous objective). This knowledge should pro-vide an important part of the theoretical and experimental basis needed to build a coastal algorithm capable of retrieving the concentration of chlorophyll and as a secondary out-put, the two other optically significant component concentrations (SPM and CDOM) from remotely sensed reflectance data.

(27)

1.4 Hypotheses

This study is based on the following hypothesis :

• Measured IOP and AOP datasets are coherent in an optical and radiative transfer sense;

The reader is referred to sections 2.1 and 2.2 for more information on IOPs, AOPs and their relationship through radiative transfer; and to section 2.4 for the definitions of optical coherence and radiative transfer closure. Validation of the research hypothesis should give us confidence about the validity of our optical dataset.

1.5 Hypothesis validation methodology

To validate the research hypotheses, the following methodology will be adopted :

1. The magnitude and the variability of measured optically-significant biogeochemi-cal parameters as well as their spatial distribution patterns will be analyzed and reported. The results will be compared with those previously reported for other coastal regions around the world.

2. The magnitude and the variability of measured IOPs as well as their spatial distri-bution patterns will be analyzed and reported. The results will be compared with those previously reported for other coastal regions around the world;

3. The IOPs will then be spectrally parameterized and the resulting coefficients will be compared with those previously reported for other coastal regions around the world;

(28)

4. The measured IOPs will be regressed against the measured optically significant bio-geochemical variables, the linearity of the (extensive) IOPs versus the concentra-tion of the corresponding constituents will help in validating the coherency of the dataset. This step will also provide the optical cross-sections (i.e. the specific IOPs), a fundamental piece of information required for remote sensing inversion models; 5. Radiative transfer simulations will be performed, taking measured bulk IOPs as an

input and simulating in-water spectral reflectance. The magnitude and shape of the computed reflectance will be compared with in-water measurements. This ison will provide a degree of closure while the process of performing these compar-isons will yield information about variables that were measured with a questionable accuracy and variables that were not measured at all.

(29)

Chapter 2 Background and theory

There are two major types of medium optical properties that can be defined in the context of remote sensing : inherent and apparent optical properties. Inherent Optical Proper-ties (IOPs) are independent of illumination conditions while Apparent Optical ProperProper-ties (AOPs) closely depend on the ambient illumination conditions. Operational instruments that enable the measurements of in-situ IOPs have only become available in the last few decades. AOPs on the other hand have a relatively longer history of field-hardened mea-surements. Below we summarize the main elements of the nomenclature and physical interpretation of IOPs, AOPs and radiative transfer theory within the context of a water medium. For a more complete description of the background theory associated with the parameters described in this section, readers are referred to Mobley (1994).

2.1 Inherent optical properties (IOPs)

There are two distinct types of IOPs : extensive (concentration dependent) and intensive (concentration independent). Intensive IOPs depend only on the physical and optical properties of a given constituent while the magnitude of extensive IOPs depend also of the number density of the specific substance they describe. The IOPs which are relevant to ocean and coastal optics are the beam absorption, attenuation and scattering coefficients as well as the volume scattering function. The spectral beam attenuation coefficient c(A)

(30)

of a medium describes the fraction of incident beam light intensity attenuated per meter of medium. The light beam is attenuated either because it was absorbed or scattered away from its direct path by the molecules in the medium. One can thus write :

c(A) = fl(A) + b(A) (2.1)

where a (A) and b{\) are the spectral beam absorption and scattering coefficients respec-tively, c, a and b are all in units of m_ 1, denoting the fraction of incident light that was

attenuated/absorbed/scattered after travelling 1 m in the water medium. The IOPs are additive, i.e. the sum of the IOPs of individual components is equal to the bulk IOP of the water column:

fl(A) = aw{\) + aph(A) + ag(k) + aNAP(A) (2.2)

= aw(\) +anw(A)

where a is the bulk absorption coefficient of the water column, aw and anw represent ab-sorption by water molecules and non-water constituents respectively, a^ is the abab-sorption coefficient of phytoplankton, ag is the absorption coefficient of CDOM and AJVAP is the

ab-sorption coefficient of non-algal particles. Among the components given above, aw is the least variable at a given wavelength (depending on the quasi-constant molecular proper-ties of seawater) and the most easily parameterized. The values of aw are usually taken from the literature (Pope and Fry, 1997). Similarly, for scattering, we can write :

b(A)=bw(\) + bp(\) (2.3)

(31)

where b is the bulk scattering coefficient of seawater, bw and bp are the scattering

coeffi-cients for water molecules and marine particles respectively. We can similarly write for the backscattering coefficient (fraction scattered at angles > 90° relative to incident beam direction):

bb(A)=bbw{A) + bbp(A) (2.4)

The volume scattering function fi(9,<p,A) describes the An steradian directional behavior of scattering with respect to the zenith angle 9 and azimuth angle (p (fraction scattered into a given direction per unit steradian per unit meter). Its integration over all directions yields b(X) :

b(A) = 2n [ j6(0,d>, A) sin{0) &9 (2.5)

Jo

while its integration over the backward hemisphere gives the backscattering coefficient

MA) :

bb(\) = In I ${9,<p, A) sin(6) d9 (2.6)

Jn/2

For pure seawater, an analytical model of $w (9, <p) is given by Morel (1974):

4.32

(2.7)

M 0 / A ) = /3W(90°, Ao) (j) [l + 0.835cos2(ct>)

bw and bbw can then be computed by numerically integrating [5W using equations 2.5 and

2.6 respectively. The volume scattering phase function fi{<p,h) is the volume scattering function normalized by the scattering coefficient:

(32)

«> - V**

(2

-

8)

where ji((p, A) is in units of sr~x. Similarly, the dimensionless backscattering ratio i>i,(b) (A), a critical parameter for remote sensing, is defined as the ratio of light scattered in the backward direction to the total scattered light:

The spectral shape of the particulate scattering coefficient is approximately modeled by a power decay function about a reference wavelength (Walker, 1994):

/ 4 4 0 \Y p

fep(A) = fcp(440) l^—j (2.10)

where Yp is the spectral slope parameter of particulate scattering coefficient. While it has often been assumed that Yp = 1 (Gordon and Morel, 1983), Babin et ah (2003a) recently observed values close to 0 over the coastal European coastal waters. The total single scat-tering albedo w is the fraction of scattered light to total attenuated light:

«(A)

=

b

M

& 1 1 )

The bulk cv and j6 parameters fully describe the inherent optical parameters of a medium in terms of a full radiative transfer description of that medium. Given the variation of these two parameters as a function of optical depth T(Z) (the product of distance from the surface and the beam attenuation coefficient T ( A , Z ) = c(A,z) * z) and the boundary

illumination conditions, one can solve the radiative transfer equation to obtain upwelling and downwelling irradiance and directional radiance fields. Similarly, the single scatter-ing albedo of particles cvv is defined as :

(33)

UrW

=m

(

,

12)

2.2 Apparent optical properties AOPs

The two AOPs that are commonly measured in the field are the spectral irradiance and the spectral radiance. The spectral irradiance E(A) is the flux of energy received on a unit horizontal area from all directions (units of W m~2nm~1). The spectral radiance L(A, O) is the directional flux of power incoming or leaving a unit area dA normal to the direction of the flux within a solid angle dCl. The radiance is and optical quantity that is directly mea-sured by remote sensing imagers and is in units of W m_ 2n m "1s r "1 where sr represents

steradians, the units of solid angle. The upwelling radiance Lu is measured by a sensor that is looking down (e.g. a satellite) while the downwelling radiance L^ is measured by a sensor that is looking up. In the water medium, one can readily measure the downwelling irradiance E^(A) which is the integration of the radiance field over the upper hemisphere

Ed(X) = / / L(X,6,(p)cos{e)sin{e)ded(p (2.13)

Jo Jo

similarly, the upwelling irradiance EU(A) is defined as :

EM(A) = / / L{A,6,d>)cos(6)sin{e)ded(p (2.14)

JO Jn/2

(34)

where Kj (A, 2) is the diffuse attenuation coefficient for the downwelling irradiance E^ (A, z) at depth z and £(j(A/0_) is the irradiance just beneath the water surface. In the context

of remote sensing, it is common practice to compute K^(\,z) relative to surface (as in eq. 2.15). If, however, E^ is known at another depth z0, Kj can be computed relative to that

depth also. Diffuse attenuation coefficients for EM(A,z), L^(A,z) and LM(A,z) can also be

computed in a similar fashion. Such coefficients can be described as "apparent" since they are approximate attenuation coefficients which nonetheless retain some dependence on the radiation field. The in-water irradiance reflectance, a dimensionless parameter, is defined as:

R

^=MtP) <

2 -

16 '

while the remote sensing reflectance is defined as :

R

^> = JJS)

(2

'

17)

where Lw(A,0+) is the water-leaving radiance just above the sea-surface. In ocean color

remote sensing, Rrs(h) is more commonly used than R(\) because satellites directly

mea-sure LM(A) instead of EM(A). If LM(A,z) is measured just below the surface with a

sub-mersible radiometer, it can be computationally transmitted through the surface interface to estimate Lro(A,0+).

2.2.1 Transmission of Lw through the ocean surface interface

The methodology followed in this section is comprehensively described in the SeaWIFS Ocean Optics Protocols (Mueller et ah, 2003). Several optical processes modify the light passing through the ocean surface interface. In order to correctly estimate Lw(\,9,(p,0+)

(35)

from the measured LM(A, 9,<p, 0~), all of these processes need to be accounted for in

in-terface transformation. To better illustrate the problem, we will first describe the fate of a sun beam entering the water surface. A portion of the downwelling light beam incident on the ocean surface at zenith angle 9 is reflected back into the atmosphere while the rest of the beam is transmitted into the water medium. The refracted beam travels in the water at a zenith angle of 9 where 9 > 9 . The two angles 9 and 9 are related through Snell's law of refraction:

sin(O) sin(9 )

where n is the index of refraction of the seawater which is usually taken as 1.34. The index of refraction of air is taken as 1. Fresnel reflectance p(0,9') is the parameter that describes what portion of the light beam is reflected back from the interface. Knowing 9 and 9 , it is possible to compute p :

Plfif) = \

sin_sin1₂ {9 - 9') tan_{(0 + 9')}₊_tan₁1_{(9+ 9')} {9 - 9') (2.19)

The incident solar angle 9 was computed using the algorithm by Michalsky (1988). Note that the water-to-air Fresnel reflectance is the parameter to consider in the case of the remote sensing problem and is always equal to its air-to-water counterpart (i.e. p(9,9 ) = p{9 ,9)) . When there are no winds and the sea surface is flat, it is assumed that the Fresnel reflectance fully explains the effects of the interface. In the more realistic case of a sea surface roughened by wind-induced waves, a model accounting for surface slope statistics (e.g. Cox and Munk 1954) is usually combined with the Fresnel reflectance. The water-leaving radiance is described as :

(36)

where the term [1 — p(9 ,6)] denotes the water-to-air Fresnel transmittance (the loss in the upwelling beam by reflection at the interface) while n2 is associated with the loss due to refraction.

2.3 The radiative transfer equation

The electromagnetic light field theory provides the radiative transfer equation (RTE), an-alytically linking the IOPs to the AOPs. The RTE can be shown in its most compact form

L{e,<p,z) d

1 +

*to

W(T) / L ( 0 ' , < / / , Z ) j g ( 0 y -> S<p) dQ! (2.21)

where L(0',<pf,z) is the incident spectral radiance in (zenith, azimuth) direction (6',(p'), L(9,cl)rz) is the spectral radiance in direction (6,(p) after being scattered in the medium, T is the optical depth, /3 is the scattering phase function, CO(T) is the single-scattering albedo and Ci' is the solid angle of the incident light beam. Currently there is no known general analytical solution to this equation which incorporates both a derivative and an angular integral of the solution being sought (i.e. the directional radiance field L). In order to estimate the AOPs from IOPs (the forward approach), one needs to numerically solve the RTE as a function of the boundary (illumination) conditions. There are a number of ways of solving the RTE. Zaneveld et al. (2005) give an overview of the theory and applications of radiative transfer theory. Mobley et al. (1993) compare the results of the concurrent runs of five different codes developed by different research teams that compute numerical solutions of the RTE for a variety of boundary conditions.

(37)

2.4 Optical coherence and radiative transfer closure

We define the optical coherence as the expected behavior of an optical parameter as a function of related water constituents and optical parameters that were measured in-dependently. For example, the backscattering coefficient by is known to be correlated with the suspended matter concentration (SPM). Likewise, a^ is supposed to be correlated with Chla concentration as well as ag is generally inversely correlated with salinity. Lack of such correlations in measured optical datasets would inevitably bring about doubts on their coherence.

Radiative transfer closure is generally defined as the agreement between measured and simulated optical parameters. Simulated AOPs are computed using radiative transfer models that employ measured (and/or modeled a n d / o r prescribed) IOPs as parametric inputs. When simulated data agree with experimental data within an acceptable range of accuracy, one can say that radiative transfer closure is achieved.

(38)

Chapter 3 Study site and data collection

3.1 St.Lawrence Gulf and Estuary

The Gulf and Estuary of St. Lawrence forms a coastal ecosystem with dimensions com-parable to those of an inland sea (Figure 3.1). The main component of the St. Lawrence ecosystem is the St.Lawrence river, connecting the Great Lakes to the Atlantic Ocean and draining out a watershed of an extremely large industrialized portion of the continental North America. The mean annual discharge of the St. Lawrence river near Quebec city is around 10.000 m3 s_ 1, accounting for more than two thirds of the total freshwater income

into the lower Estuary (Ingram and El-Sabh, 1990). The St. Lawrence is a very important seaway, with 5.106 tons of crude oil being transported each year (as of 1991). The second

most important freshwater source is the Saguenay River which mixes with the St. Lawrence near Tadoussac. The Laurentian channel is a large and deep bathymetric feature that extends from the Atlantic Ocean towards the Estuary u p to near Tadoussac. The western head of the channel is characterized by a sharp bathymetric gradient offshore Tadoussac where the channel depth that exceeds300 m over the entire Gulf and maritime Estuary rapidly de-creases to a few tens of meters. This feature creates a nearly permanent tidally-induced

(39)

upwelling of cold, nutrient-rich North-Atlantic waters, increasing the biological activity of the region (Therriault et al, 1990). Several other bathymetric features are responsible for the local sharp gradients that are frequently observed in the St. Lawrence (Ingram and El-Sabh, 1990). Other important freshwater sources are the Manicougan and Aux-Outardes rivers that feed into the lower St. Lawrence Estuary (area extending from Tadonssac to Pointe-des-Monts). The Anticosti Gyre is a wind and Coriolis-force driven vortex-like cy-clonic surface current and is located between the western tip of Anticosti Island, Gaspe Current and Pointe-des-Monts.

There is a positive salinity gradient from west to east, observable throughout the year. Seawater thus penetrates inland while freshwater moves out towards the sea in the surface layer.

Figure 3.1: The main geographical features of the Gulf and Estuary of St. Lawrence. The main map was taken from Koutitonsky and Bugden (1991), the bathymetry map was taken from Sundby and Bewers (1991).

The physical parameters of the ecosystem are known to exhibit an extremely high spa-tial and temporal variability as compared to similar ecosystems around the world. These conditions induce complex primary productivity patterns (Le Fouest et al, 2005).

(40)

Therri-ault and Levasseur (1985) classified the St. Lawrence Estuary into four sub-regions that are distinct from one another as far as their primary productivity patterns are concerned (Figure 3.2) :

The outflow region : Turbid riverine waters are rapidly flushed towards the Gulf along

the south shore by Coriolis-force dominated currents, creating the Gaspe current. The eastern extent of the current is controlled by the seasonal flowrates of the St. Lawrence (Ingram and El-Sabh, 1990), but usually penetrates into the Gulf. The productivity on the south shore is low because of low residence times and low light levels.

The upwelling region : High nutrient loads and low temperatures due to tidally-induced

upwellings at the head of the Laurentian channel dominate this region. A higher productivity is usually observed in this area.

The plume region : This area is under the constant influence of the Manicougan and

Aux-Outardes river plumes. Continuous nutrient replenishment and stabilization regimes ensure a higher production than in the upwelling region.

The near-Gulf region : The near-Gulf region exhibits the least turbid waters and the

deepest photic layer of the lower St. Lawrence Estuary. This region is vertically most stable, with low nutrients and high temperatures. It is the least influenced by tides and is a more Gulf-like regime with nutrient-limited production in summer and sub-surface chlorophyll maxima.

As briefly discussed above, freshwater inputs from different sources, various bathymetric features, tides etc. make the horizontal and vertical mixing patterns difficult to predict (Le Fouest et ah, 2006). These particular conditions result in extremely high variations of the spatial and temporal distributions of biogeochemical variables. In order to have a better understanding of the role of this particular ecosystem in global primary production

(41)

Figure 3.2: The four distinct primary production zones of the lower St. Lawrence estuary as defined by Therriault et al. (1990). Illustration adapted from the work of the author. and climate change, bio-optical variables that are indicative of biogeochemical variables need to be extensively monitored using timeseries of satellite imagery. To better exploit remotely-sensed data acquired over such a complex coastal environment, one needs to understand the relationships between the biogeochemical and bio-optical variables (see section 1.1). The St.Lawrence bio-optical monitoring program referred to above was the first attempt to acquire over the St. Lawrence an extensive multi-year and multi-season temporal series of concurrently measured biogeochemical and bio-optical variables.

3.1.1 Bio-optical cruises

Figure 3.3 shows the location of all the stations that were visited during the five two-weeks cruises that took place between 1997-2001. Various collaborators were involved in the project, helping in the acquisition of the different parts of the resulting dataset. The

(42)

IOP data were mainly collected during the two cruises that took place in May 2000 and April 2001. The ensemble of collected IOP data are summarized in Table 3.1. The data collection and validation protocols of SeaWiFS (Mueller et ah, 2003) and JGOFS (JGOFS, 1991) were followed during all five cruises.

m"N't

I

«f w 62" W 5fl W

48"N •*r\

4SU _{• too xamm}

Figure 3.3: Locations of all the stations that were visited during the 1997-2001 cruises.

Table 3.1: Inherent optical parameters measured in the Estuary and Gulf of St.Lawrence. The number of stations visited is given between parenthesis.

flnw(A)fl Cnw\A-)

a

g

(\y

ag(A)d ap{X)e aph(A)e flNAp(A)e j5(0,A/ Aug 1997 28-Sept 12 (39) X 1998 Oct 21-Nov 3 (22) X 1999 Jun 27-Jul 6 (38) X 2000 May 18-Jun 3 (42) X" Xb X X X X 2001 Apr 20-May 4 (33) X X X X X X X X

flac-9 measurements without a 0.2 }im filter. bEstuary only.

cac-9 measurements with a 0.2 /im filter.

dSpectrophotometric CDOM absorption measurements.

eParticulate absorption measurements with the filter-pad technique.

^Measurements of the volume scattering function at 3 backward angles (100°7125o,150°) and two wavelengths (450 and 530 nm) with a WET Labs

ECO-VSF.

(43)

3.2 In-water constituents of optical relevance

Concentrations of several in-water constituents were determined from water samples col-lected from three depths (surface, 10% light and the maximum chlorophyll depth). Pig-ments (including the Phaeopigment Phaeo, an indicator of metabolism by-products; hence the age of the bloom) were determined with a high-performance liquid chromatography (HPLC) method (JGOFS, 1991). Chla was also determined with the traditional fluorimetric method (Strickland and Parsons, 1972). Total suspended matter (SPM) concentration was determined by filtering water collected from the surface, 51%, 10%, 1%, 0,1% light depths as well as from 60 m and from the maximum-chlorophyll depth using 0.4 \im polycarbon-ate filters. Particulpolycarbon-ate Inorganic Matter (PIM) was determined (except in September) by filtering water samples through Whatman G F / F glass fiber filters (mean diameter of 0.7 }im) and weighting the filters before and after combustion to eliminate the organic matter. Particulate Organic Matter (POM) and the percent fraction of organic particles (%POM) were deduced from:

POM = SPM - PIM (3.1) and

POM

°/oPOM = j ^ (3.2)

Due to the difficulties associated with the measurement methods (Carder et al., 1989), CDOM concentration was not measured in situ. Instead, CDOM absorption coefficient Ag(A) was used as a proxy for CDOM concentration. The reader is referred to sections 3.3.1 and 3.3.2 for more details about ag(\) measurements. At each station, vertical profiles of temperature, salinity and conductivity were recorded with a SeaBird CTD. Concentrations of nutrients, particulate organic nitrogen and particulate organic carbon were measured. Phytoplankton identification was done by microscopy. These measurements were not

(44)

presented in the current study.

3.3 Inherent optical properties

3.3.1 Absorption and scattering with the ac-9

Instrument description

In-situ vertical profiles of the total absorption and attenuation coefficients were performed with a WET Labs ac-9 in-situ spectrophotometer (Zaneveld et al., 1992). The ac-9 is made of two 25 cm-long flow tubes. A polychromatic light beam is generated at the entrance of each tube and the light beam intensity exiting the tube is measured with a detector. Nine filters mounted on a rotating wheel filter the light into 9 wavelength channels (412, 440,488, 510, 532,555, 650, 676, 715 nm). The internal surface of one of the tubes is made of a material which reflects most of the scattered light back into the field of view of the sensor located at the end of the tube. The difference between the light intensity entering and leaving the tube is therefore due to absorption only. Absorption data must later be corrected for the residual light not reaching the detector due to scattering of the photons away from the field of view of the sensor. The second tube has absorbing inner walls so that out-scattered light does not enter the field of view of the detector. This measurement thus yields an estimation of the attenuation coefficient. The ac-9 takes continuous mea-surements at 6 Hz and is able to resolve fine-scale vertical variability regardless of the ambient illumination conditions. The absorption measurements are performed relative to pure seawater i.e. after post-processing, the instrument provides anw(X) and cnw(X). Ignoring the small scattering contribution due to dissolved matter, vertical profiles of the particulate scattering coefficient bp(\) can be derived from these measurements using :

(45)

&P(A) = cnH,(A) - awa,(A) (3.3)

The particle absorption coefficient ap(\) is computed from :

flp(A) = aKH,(A) - flg(A) (3.4)

cig is described in section 3.3.1. The total absorption coefficient was computed using :

a(A) = anw{\) + aw{\) (3.5)

where the subscript w stands for pure seawater. Bulk scattering and attenuation coef-ficients were also determined in a similar way. The reader is referred to section 2.1 for more details about the absorption and scattering properties of pure seawater. The mass-specific particulate scattering coefficient b* is in units of m2.g_ 1 and can be computed via

a normalization by SPM (Babin et ah, 2003a):

m = ^

0.6)

In contrast with a^AP (equation 3.15), b* includes the optical characteristics of phytoplank-ton particles (cells).

Correction for the instrument drift with time

The ac-9 is known to drift with time. In order to account for the drift, measurements with deionized water were performed each day of the cruise. This calibration water was instantly prepared onboard with a Barnstead NANOpure water purification unit. The water

(46)

calibration procedure followed is the one described in the WET Labs ac-9 data processing guide (Wetlabs, 2000). Drift correction mainly consists of subtracting deionized water readings from that day's ac-9 data. Results of the water calibration are given in Figure 3.4. The improvement can be characterized in terms of the mean and rms spread of the curves before correction (a) and after correction (b).

0.02 0.02 -0.02 -0.04 -0.06 -0.08 400 500 600 700 Wavelength [nm] 800 -0.1 400 500 600 700 800 Wavelength [nm]

Figure 3.4: The effects of the ac-9 drift correction on deionized water absorption spectra : (a) before the correction, (b) after the correction. Shown above are all individual ac-9 measurements performed with deionized water (the April 2001 cruise).

Temperature and salinity corrections

The absorption coefficient of the water is known to linearly depend on the temperature in the near-infrared region (Pegau and Zaneveld, 1993). The drop in the absorption data at 715 nm that is observed in Figure 3.4 is predominantly due to the fact that the tem-perature of the deionized water used in the calibration of the instrument in the field and the temperature of the water used in the calibration of the instrument in the factory were

(47)

different. The correction to be applied to the absorption data is made according to the following formula:

anw = [anw]m ~ [ft * {t ~ tr) + Ys * (S - Sr)\ (3.7)

where anw is the corrected absorption coefficient, [anw]m is the measured absorption co-efficient, Yf and Ys are the slopes of seawater dependence of temperature and salinity

respectively. The parameters t and S are the temperature and salinity of the water that is being sampled and they have to be measured during the calibration session. tr and Sr are the reference temperature and salinity at which Yf, Ys were determined. Yf, Ys, tr and Sr were given by the manufacturer. Similarly, the corrections for the c tube were computed using the following formula :

Cnw = [Cnw]m ~ [Ys * ( S - Sr)] (3.8)

Figure 3.5 illustrates the effects of the ac-9 temperature and salinity corrections on drift-corrected deionized water spectra. The magnitude of salinity correction is very small compared with temperature correction which is especially significant at 715 nm. One can note significant improvement for the wavelengths about 700 nm while the correction does little to reduce the dispersion at shorter wavelengths.

Scattering correction

The employed correction for residual scattering (due to those out-scattered photons which are not redirected into the detector field-of-view (FOV) by the reflecting walls) consists of subtracting a(715) from all wavelengths. This method is commonly used by the ocean optics community (Babin et ah, 2003a) and assumes that the absorption by particles and

(48)

0.01 -0.06 U.U 1 0 0.01 0.02 0.03 0.04 0.05 ~ ~" / -• W -400 500 600 700 800 Wavelength [nm] 400 500 600 700 800 Wavelength [nm]

Figure 3.5: The effects of the ac-9 temperature correction on deionized water absorption spectra : (a) before the correction, (b) after the correction. Shown above are all individual ac-9 measurements performed with deionized water (the April 2001 cruise).

CDOM can both be neglected at that wavelength and that the volume scattering coeffi-cient (over the angles where out-scattering away from the detector FOV occurs) is inde-pendent of wavelength (Wetlabs, 2000). The effects of the scattering correction on pure-water measurements are illustrated in Figure 3.6. It is worth noting that after applying all the corrections, the precision of the instrument (as represented by the spread of the differ-ent curves in Figure 3.6) reaches the manufacturer's specification that is q=0.005 m_ 1. Since

absorption by water is not included in the measurements and since we can safely assume that no particles can get through the filter (see next section), the scattering correction was not applied to ag measurements (Zaneveld et ah, 1992).

Absorption by CDOM with the ac-9

The ac-9 was also operated with a Gellman 0.2 \im Maxi Capsule filter attached into the intake hose of the ac-9 (in April 2001 only), yielding absorption by CDOM ag(\)

(49)

0.01 0.005

I

_-0.005 -0.01 -0.015 -0.02 "_ -.^~-" -z-" _,.,/ i j 1 < \\ v - 7 / Aj --'-^h | \ J (a) 0.01 0.005 -0.005 -0.01 -0.015 400 500 600 700 800 Wavelength [nm] -0.02 400 500 600 700 800 Wavelength [nm]

Figure 3.6: The effects of the ac-9 scattering correction on deionized water absorption spectra (already corrected for instrument's drift and temperature and salinity effects) : (a) before the correction, (b) after the correction. Shown above are all individual ac-9 measurements performed with deionized water (the April 2001 cruise).

dowski et ah (1999). When the filter was present, the water flowrate was significantly reduced (0.8 L/min) in contrast with the usual flowrate without filter (5L/min). Due to this very low flowrate, it was not possible to realize continuous profiles with the filter. Instead, the instrument was held at several intermittent depths near surface until all the water in the tubes was purged. The CDOM absorption data were corrected for instrument drift and salinity and temperature effects (see above). To empirically estimate the spectral slope coefficient Sg, resulting spectra were fitted to the equation :

a

g

{\) = a

g

(\

ref

)

e

-%(

A

-V)

(3.9) where \ref = 440 nm. It was the fitted ag spectra that was used in the computation of av in equation 3.4. Use of fitted data rather than actual measurements helped in reducing error, having a smooth CDOM spectra in the near-infrared channels; it reduced the noise that

(50)

is apparent at these wavelengths due to the low value of the absorption coefficient. The resulting value of ap(715) was always within the instrument's precision bounds around

zero (-0.005 < ap(715) < O.OOSm^1). Note that ap(715) was subtracted from all channels

as part of the flp(A) calculation in order to be coherent with the assumptions of the

scat-tering error correction employed in the previous section, otherwise one obtains negative flp values at 715 nm and sometimes, at 676 nm.

3.3.2 Spectrophotometric CDOM absorption

Spectrophotometer measurements of ag made onboard the research vessel were the only IOPs that were measured over the full 5 year period of our field campaigns. The instru-ment used was a Lambda-6 spectrophotometer with a working precision of ±0.05 m_ 1.

Data collected with the spectrophotometer was noisy in the near-infrared channels due to the very low magnitude of the CDOM absorption coefficient in this spectral region. A baseline correction was applied to the data by subtracting the mean value of ag across the 683-688 nm range from all wavelengths as described in Babin et ah (2003b). The resulting data were then fitted to the traditional spectral shape function given in equation 3.9. In-stead of performing a first-order linear regression on log-transformed data, the Matlab im-plementation of the Nelder-Mead non-linear minimization was used, giving more weight to low-wavelength measurements that have higher magnitudes (Stedmon and Markager, 2001). It is known that Sg determined by spectrally fitting empirical data depends on the spectral range used in the fit (Carder et ah, 1989, Twardowski et ah, 2004). In our case, we only used data from the 350-500 nm range because the measurements at longer wave-lengths had very low magnitudes that are close to the instrument sensitivity limit.